]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - drivers/net/geneve.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
net: phy: phylink: fix uninitialized variable in phylink_get_mac_state
[mirror_ubuntu-bionic-kernel.git] / drivers / net / geneve.c
1 /*
2 * GENEVE: Generic Network Virtualization Encapsulation
3 *
4 * Copyright (c) 2015 Red Hat, Inc.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
9 */
10
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12
13 #include <linux/kernel.h>
14 #include <linux/module.h>
15 #include <linux/etherdevice.h>
16 #include <linux/hash.h>
17 #include <net/dst_metadata.h>
18 #include <net/gro_cells.h>
19 #include <net/rtnetlink.h>
20 #include <net/geneve.h>
21 #include <net/protocol.h>
22
23 #define GENEVE_NETDEV_VER "0.6"
24
25 #define GENEVE_UDP_PORT 6081
26
27 #define GENEVE_N_VID (1u << 24)
28 #define GENEVE_VID_MASK (GENEVE_N_VID - 1)
29
30 #define VNI_HASH_BITS 10
31 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
32
33 static bool log_ecn_error = true;
34 module_param(log_ecn_error, bool, 0644);
35 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
36
37 #define GENEVE_VER 0
38 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
39
40 /* per-network namespace private data for this module */
41 struct geneve_net {
42 struct list_head geneve_list;
43 struct list_head sock_list;
44 };
45
46 static unsigned int geneve_net_id;
47
48 struct geneve_dev_node {
49 struct hlist_node hlist;
50 struct geneve_dev *geneve;
51 };
52
53 /* Pseudo network device */
54 struct geneve_dev {
55 struct geneve_dev_node hlist4; /* vni hash table for IPv4 socket */
56 #if IS_ENABLED(CONFIG_IPV6)
57 struct geneve_dev_node hlist6; /* vni hash table for IPv6 socket */
58 #endif
59 struct net *net; /* netns for packet i/o */
60 struct net_device *dev; /* netdev for geneve tunnel */
61 struct ip_tunnel_info info;
62 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */
63 #if IS_ENABLED(CONFIG_IPV6)
64 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */
65 #endif
66 struct list_head next; /* geneve's per namespace list */
67 struct gro_cells gro_cells;
68 bool collect_md;
69 bool use_udp6_rx_checksums;
70 };
71
72 struct geneve_sock {
73 bool collect_md;
74 struct list_head list;
75 struct socket *sock;
76 struct rcu_head rcu;
77 int refcnt;
78 struct hlist_head vni_list[VNI_HASH_SIZE];
79 };
80
81 static inline __u32 geneve_net_vni_hash(u8 vni[3])
82 {
83 __u32 vnid;
84
85 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
86 return hash_32(vnid, VNI_HASH_BITS);
87 }
88
89 static __be64 vni_to_tunnel_id(const __u8 *vni)
90 {
91 #ifdef __BIG_ENDIAN
92 return (vni[0] << 16) | (vni[1] << 8) | vni[2];
93 #else
94 return (__force __be64)(((__force u64)vni[0] << 40) |
95 ((__force u64)vni[1] << 48) |
96 ((__force u64)vni[2] << 56));
97 #endif
98 }
99
100 /* Convert 64 bit tunnel ID to 24 bit VNI. */
101 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
102 {
103 #ifdef __BIG_ENDIAN
104 vni[0] = (__force __u8)(tun_id >> 16);
105 vni[1] = (__force __u8)(tun_id >> 8);
106 vni[2] = (__force __u8)tun_id;
107 #else
108 vni[0] = (__force __u8)((__force u64)tun_id >> 40);
109 vni[1] = (__force __u8)((__force u64)tun_id >> 48);
110 vni[2] = (__force __u8)((__force u64)tun_id >> 56);
111 #endif
112 }
113
114 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
115 {
116 return !memcmp(vni, &tun_id[5], 3);
117 }
118
119 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
120 {
121 return gs->sock->sk->sk_family;
122 }
123
124 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
125 __be32 addr, u8 vni[])
126 {
127 struct hlist_head *vni_list_head;
128 struct geneve_dev_node *node;
129 __u32 hash;
130
131 /* Find the device for this VNI */
132 hash = geneve_net_vni_hash(vni);
133 vni_list_head = &gs->vni_list[hash];
134 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
135 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) &&
136 addr == node->geneve->info.key.u.ipv4.dst)
137 return node->geneve;
138 }
139 return NULL;
140 }
141
142 #if IS_ENABLED(CONFIG_IPV6)
143 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
144 struct in6_addr addr6, u8 vni[])
145 {
146 struct hlist_head *vni_list_head;
147 struct geneve_dev_node *node;
148 __u32 hash;
149
150 /* Find the device for this VNI */
151 hash = geneve_net_vni_hash(vni);
152 vni_list_head = &gs->vni_list[hash];
153 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
154 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) &&
155 ipv6_addr_equal(&addr6, &node->geneve->info.key.u.ipv6.dst))
156 return node->geneve;
157 }
158 return NULL;
159 }
160 #endif
161
162 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
163 {
164 return (struct genevehdr *)(udp_hdr(skb) + 1);
165 }
166
167 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
168 struct sk_buff *skb)
169 {
170 static u8 zero_vni[3];
171 u8 *vni;
172
173 if (geneve_get_sk_family(gs) == AF_INET) {
174 struct iphdr *iph;
175 __be32 addr;
176
177 iph = ip_hdr(skb); /* outer IP header... */
178
179 if (gs->collect_md) {
180 vni = zero_vni;
181 addr = 0;
182 } else {
183 vni = geneve_hdr(skb)->vni;
184 addr = iph->saddr;
185 }
186
187 return geneve_lookup(gs, addr, vni);
188 #if IS_ENABLED(CONFIG_IPV6)
189 } else if (geneve_get_sk_family(gs) == AF_INET6) {
190 static struct in6_addr zero_addr6;
191 struct ipv6hdr *ip6h;
192 struct in6_addr addr6;
193
194 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
195
196 if (gs->collect_md) {
197 vni = zero_vni;
198 addr6 = zero_addr6;
199 } else {
200 vni = geneve_hdr(skb)->vni;
201 addr6 = ip6h->saddr;
202 }
203
204 return geneve6_lookup(gs, addr6, vni);
205 #endif
206 }
207 return NULL;
208 }
209
210 /* geneve receive/decap routine */
211 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
212 struct sk_buff *skb)
213 {
214 struct genevehdr *gnvh = geneve_hdr(skb);
215 struct metadata_dst *tun_dst = NULL;
216 struct pcpu_sw_netstats *stats;
217 unsigned int len;
218 int err = 0;
219 void *oiph;
220
221 if (ip_tunnel_collect_metadata() || gs->collect_md) {
222 __be16 flags;
223
224 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
225 (gnvh->oam ? TUNNEL_OAM : 0) |
226 (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
227
228 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
229 vni_to_tunnel_id(gnvh->vni),
230 gnvh->opt_len * 4);
231 if (!tun_dst) {
232 geneve->dev->stats.rx_dropped++;
233 goto drop;
234 }
235 /* Update tunnel dst according to Geneve options. */
236 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
237 gnvh->options, gnvh->opt_len * 4);
238 } else {
239 /* Drop packets w/ critical options,
240 * since we don't support any...
241 */
242 if (gnvh->critical) {
243 geneve->dev->stats.rx_frame_errors++;
244 geneve->dev->stats.rx_errors++;
245 goto drop;
246 }
247 }
248
249 skb_reset_mac_header(skb);
250 skb->protocol = eth_type_trans(skb, geneve->dev);
251 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
252
253 if (tun_dst)
254 skb_dst_set(skb, &tun_dst->dst);
255
256 /* Ignore packet loops (and multicast echo) */
257 if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
258 geneve->dev->stats.rx_errors++;
259 goto drop;
260 }
261
262 oiph = skb_network_header(skb);
263 skb_reset_network_header(skb);
264
265 if (geneve_get_sk_family(gs) == AF_INET)
266 err = IP_ECN_decapsulate(oiph, skb);
267 #if IS_ENABLED(CONFIG_IPV6)
268 else
269 err = IP6_ECN_decapsulate(oiph, skb);
270 #endif
271
272 if (unlikely(err)) {
273 if (log_ecn_error) {
274 if (geneve_get_sk_family(gs) == AF_INET)
275 net_info_ratelimited("non-ECT from %pI4 "
276 "with TOS=%#x\n",
277 &((struct iphdr *)oiph)->saddr,
278 ((struct iphdr *)oiph)->tos);
279 #if IS_ENABLED(CONFIG_IPV6)
280 else
281 net_info_ratelimited("non-ECT from %pI6\n",
282 &((struct ipv6hdr *)oiph)->saddr);
283 #endif
284 }
285 if (err > 1) {
286 ++geneve->dev->stats.rx_frame_errors;
287 ++geneve->dev->stats.rx_errors;
288 goto drop;
289 }
290 }
291
292 len = skb->len;
293 err = gro_cells_receive(&geneve->gro_cells, skb);
294 if (likely(err == NET_RX_SUCCESS)) {
295 stats = this_cpu_ptr(geneve->dev->tstats);
296 u64_stats_update_begin(&stats->syncp);
297 stats->rx_packets++;
298 stats->rx_bytes += len;
299 u64_stats_update_end(&stats->syncp);
300 }
301 return;
302 drop:
303 /* Consume bad packet */
304 kfree_skb(skb);
305 }
306
307 /* Setup stats when device is created */
308 static int geneve_init(struct net_device *dev)
309 {
310 struct geneve_dev *geneve = netdev_priv(dev);
311 int err;
312
313 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
314 if (!dev->tstats)
315 return -ENOMEM;
316
317 err = gro_cells_init(&geneve->gro_cells, dev);
318 if (err) {
319 free_percpu(dev->tstats);
320 return err;
321 }
322
323 err = dst_cache_init(&geneve->info.dst_cache, GFP_KERNEL);
324 if (err) {
325 free_percpu(dev->tstats);
326 gro_cells_destroy(&geneve->gro_cells);
327 return err;
328 }
329 return 0;
330 }
331
332 static void geneve_uninit(struct net_device *dev)
333 {
334 struct geneve_dev *geneve = netdev_priv(dev);
335
336 dst_cache_destroy(&geneve->info.dst_cache);
337 gro_cells_destroy(&geneve->gro_cells);
338 free_percpu(dev->tstats);
339 }
340
341 /* Callback from net/ipv4/udp.c to receive packets */
342 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
343 {
344 struct genevehdr *geneveh;
345 struct geneve_dev *geneve;
346 struct geneve_sock *gs;
347 int opts_len;
348
349 /* Need UDP and Geneve header to be present */
350 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
351 goto drop;
352
353 /* Return packets with reserved bits set */
354 geneveh = geneve_hdr(skb);
355 if (unlikely(geneveh->ver != GENEVE_VER))
356 goto drop;
357
358 if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
359 goto drop;
360
361 gs = rcu_dereference_sk_user_data(sk);
362 if (!gs)
363 goto drop;
364
365 geneve = geneve_lookup_skb(gs, skb);
366 if (!geneve)
367 goto drop;
368
369 opts_len = geneveh->opt_len * 4;
370 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
371 htons(ETH_P_TEB),
372 !net_eq(geneve->net, dev_net(geneve->dev)))) {
373 geneve->dev->stats.rx_dropped++;
374 goto drop;
375 }
376
377 geneve_rx(geneve, gs, skb);
378 return 0;
379
380 drop:
381 /* Consume bad packet */
382 kfree_skb(skb);
383 return 0;
384 }
385
386 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
387 __be16 port, bool ipv6_rx_csum)
388 {
389 struct socket *sock;
390 struct udp_port_cfg udp_conf;
391 int err;
392
393 memset(&udp_conf, 0, sizeof(udp_conf));
394
395 if (ipv6) {
396 udp_conf.family = AF_INET6;
397 udp_conf.ipv6_v6only = 1;
398 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
399 } else {
400 udp_conf.family = AF_INET;
401 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
402 }
403
404 udp_conf.local_udp_port = port;
405
406 /* Open UDP socket */
407 err = udp_sock_create(net, &udp_conf, &sock);
408 if (err < 0)
409 return ERR_PTR(err);
410
411 return sock;
412 }
413
414 static int geneve_hlen(struct genevehdr *gh)
415 {
416 return sizeof(*gh) + gh->opt_len * 4;
417 }
418
419 static struct sk_buff **geneve_gro_receive(struct sock *sk,
420 struct sk_buff **head,
421 struct sk_buff *skb)
422 {
423 struct sk_buff *p, **pp = NULL;
424 struct genevehdr *gh, *gh2;
425 unsigned int hlen, gh_len, off_gnv;
426 const struct packet_offload *ptype;
427 __be16 type;
428 int flush = 1;
429
430 off_gnv = skb_gro_offset(skb);
431 hlen = off_gnv + sizeof(*gh);
432 gh = skb_gro_header_fast(skb, off_gnv);
433 if (skb_gro_header_hard(skb, hlen)) {
434 gh = skb_gro_header_slow(skb, hlen, off_gnv);
435 if (unlikely(!gh))
436 goto out;
437 }
438
439 if (gh->ver != GENEVE_VER || gh->oam)
440 goto out;
441 gh_len = geneve_hlen(gh);
442
443 hlen = off_gnv + gh_len;
444 if (skb_gro_header_hard(skb, hlen)) {
445 gh = skb_gro_header_slow(skb, hlen, off_gnv);
446 if (unlikely(!gh))
447 goto out;
448 }
449
450 for (p = *head; p; p = p->next) {
451 if (!NAPI_GRO_CB(p)->same_flow)
452 continue;
453
454 gh2 = (struct genevehdr *)(p->data + off_gnv);
455 if (gh->opt_len != gh2->opt_len ||
456 memcmp(gh, gh2, gh_len)) {
457 NAPI_GRO_CB(p)->same_flow = 0;
458 continue;
459 }
460 }
461
462 type = gh->proto_type;
463
464 rcu_read_lock();
465 ptype = gro_find_receive_by_type(type);
466 if (!ptype)
467 goto out_unlock;
468
469 skb_gro_pull(skb, gh_len);
470 skb_gro_postpull_rcsum(skb, gh, gh_len);
471 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
472 flush = 0;
473
474 out_unlock:
475 rcu_read_unlock();
476 out:
477 skb_gro_flush_final(skb, pp, flush);
478
479 return pp;
480 }
481
482 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
483 int nhoff)
484 {
485 struct genevehdr *gh;
486 struct packet_offload *ptype;
487 __be16 type;
488 int gh_len;
489 int err = -ENOSYS;
490
491 gh = (struct genevehdr *)(skb->data + nhoff);
492 gh_len = geneve_hlen(gh);
493 type = gh->proto_type;
494
495 rcu_read_lock();
496 ptype = gro_find_complete_by_type(type);
497 if (ptype)
498 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
499
500 rcu_read_unlock();
501
502 skb_set_inner_mac_header(skb, nhoff + gh_len);
503
504 return err;
505 }
506
507 /* Create new listen socket if needed */
508 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
509 bool ipv6, bool ipv6_rx_csum)
510 {
511 struct geneve_net *gn = net_generic(net, geneve_net_id);
512 struct geneve_sock *gs;
513 struct socket *sock;
514 struct udp_tunnel_sock_cfg tunnel_cfg;
515 int h;
516
517 gs = kzalloc(sizeof(*gs), GFP_KERNEL);
518 if (!gs)
519 return ERR_PTR(-ENOMEM);
520
521 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
522 if (IS_ERR(sock)) {
523 kfree(gs);
524 return ERR_CAST(sock);
525 }
526
527 gs->sock = sock;
528 gs->refcnt = 1;
529 for (h = 0; h < VNI_HASH_SIZE; ++h)
530 INIT_HLIST_HEAD(&gs->vni_list[h]);
531
532 /* Initialize the geneve udp offloads structure */
533 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
534
535 /* Mark socket as an encapsulation socket */
536 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
537 tunnel_cfg.sk_user_data = gs;
538 tunnel_cfg.encap_type = 1;
539 tunnel_cfg.gro_receive = geneve_gro_receive;
540 tunnel_cfg.gro_complete = geneve_gro_complete;
541 tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
542 tunnel_cfg.encap_destroy = NULL;
543 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
544 list_add(&gs->list, &gn->sock_list);
545 return gs;
546 }
547
548 static void __geneve_sock_release(struct geneve_sock *gs)
549 {
550 if (!gs || --gs->refcnt)
551 return;
552
553 list_del(&gs->list);
554 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
555 udp_tunnel_sock_release(gs->sock);
556 kfree_rcu(gs, rcu);
557 }
558
559 static void geneve_sock_release(struct geneve_dev *geneve)
560 {
561 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
562 #if IS_ENABLED(CONFIG_IPV6)
563 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
564
565 rcu_assign_pointer(geneve->sock6, NULL);
566 #endif
567
568 rcu_assign_pointer(geneve->sock4, NULL);
569 synchronize_net();
570
571 __geneve_sock_release(gs4);
572 #if IS_ENABLED(CONFIG_IPV6)
573 __geneve_sock_release(gs6);
574 #endif
575 }
576
577 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
578 sa_family_t family,
579 __be16 dst_port)
580 {
581 struct geneve_sock *gs;
582
583 list_for_each_entry(gs, &gn->sock_list, list) {
584 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
585 geneve_get_sk_family(gs) == family) {
586 return gs;
587 }
588 }
589 return NULL;
590 }
591
592 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
593 {
594 struct net *net = geneve->net;
595 struct geneve_net *gn = net_generic(net, geneve_net_id);
596 struct geneve_dev_node *node;
597 struct geneve_sock *gs;
598 __u8 vni[3];
599 __u32 hash;
600
601 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->info.key.tp_dst);
602 if (gs) {
603 gs->refcnt++;
604 goto out;
605 }
606
607 gs = geneve_socket_create(net, geneve->info.key.tp_dst, ipv6,
608 geneve->use_udp6_rx_checksums);
609 if (IS_ERR(gs))
610 return PTR_ERR(gs);
611
612 out:
613 gs->collect_md = geneve->collect_md;
614 #if IS_ENABLED(CONFIG_IPV6)
615 if (ipv6) {
616 rcu_assign_pointer(geneve->sock6, gs);
617 node = &geneve->hlist6;
618 } else
619 #endif
620 {
621 rcu_assign_pointer(geneve->sock4, gs);
622 node = &geneve->hlist4;
623 }
624 node->geneve = geneve;
625
626 tunnel_id_to_vni(geneve->info.key.tun_id, vni);
627 hash = geneve_net_vni_hash(vni);
628 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
629 return 0;
630 }
631
632 static int geneve_open(struct net_device *dev)
633 {
634 struct geneve_dev *geneve = netdev_priv(dev);
635 bool metadata = geneve->collect_md;
636 bool ipv4, ipv6;
637 int ret = 0;
638
639 ipv6 = geneve->info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
640 ipv4 = !ipv6 || metadata;
641 #if IS_ENABLED(CONFIG_IPV6)
642 if (ipv6) {
643 ret = geneve_sock_add(geneve, true);
644 if (ret < 0 && ret != -EAFNOSUPPORT)
645 ipv4 = false;
646 }
647 #endif
648 if (ipv4)
649 ret = geneve_sock_add(geneve, false);
650 if (ret < 0)
651 geneve_sock_release(geneve);
652
653 return ret;
654 }
655
656 static int geneve_stop(struct net_device *dev)
657 {
658 struct geneve_dev *geneve = netdev_priv(dev);
659
660 hlist_del_init_rcu(&geneve->hlist4.hlist);
661 #if IS_ENABLED(CONFIG_IPV6)
662 hlist_del_init_rcu(&geneve->hlist6.hlist);
663 #endif
664 geneve_sock_release(geneve);
665 return 0;
666 }
667
668 static void geneve_build_header(struct genevehdr *geneveh,
669 const struct ip_tunnel_info *info)
670 {
671 geneveh->ver = GENEVE_VER;
672 geneveh->opt_len = info->options_len / 4;
673 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
674 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
675 geneveh->rsvd1 = 0;
676 tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
677 geneveh->proto_type = htons(ETH_P_TEB);
678 geneveh->rsvd2 = 0;
679
680 ip_tunnel_info_opts_get(geneveh->options, info);
681 }
682
683 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
684 const struct ip_tunnel_info *info,
685 bool xnet, int ip_hdr_len)
686 {
687 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
688 struct genevehdr *gnvh;
689 int min_headroom;
690 int err;
691
692 skb_reset_mac_header(skb);
693 skb_scrub_packet(skb, xnet);
694
695 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
696 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
697 err = skb_cow_head(skb, min_headroom);
698 if (unlikely(err))
699 goto free_dst;
700
701 err = udp_tunnel_handle_offloads(skb, udp_sum);
702 if (err)
703 goto free_dst;
704
705 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
706 geneve_build_header(gnvh, info);
707 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
708 return 0;
709
710 free_dst:
711 dst_release(dst);
712 return err;
713 }
714
715 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
716 struct net_device *dev,
717 struct geneve_sock *gs4,
718 struct flowi4 *fl4,
719 const struct ip_tunnel_info *info)
720 {
721 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
722 struct geneve_dev *geneve = netdev_priv(dev);
723 struct dst_cache *dst_cache;
724 struct rtable *rt = NULL;
725 __u8 tos;
726
727 if (!gs4)
728 return ERR_PTR(-EIO);
729
730 memset(fl4, 0, sizeof(*fl4));
731 fl4->flowi4_mark = skb->mark;
732 fl4->flowi4_proto = IPPROTO_UDP;
733 fl4->daddr = info->key.u.ipv4.dst;
734 fl4->saddr = info->key.u.ipv4.src;
735
736 tos = info->key.tos;
737 if ((tos == 1) && !geneve->collect_md) {
738 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
739 use_cache = false;
740 }
741 fl4->flowi4_tos = RT_TOS(tos);
742
743 dst_cache = (struct dst_cache *)&info->dst_cache;
744 if (use_cache) {
745 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
746 if (rt)
747 return rt;
748 }
749 rt = ip_route_output_key(geneve->net, fl4);
750 if (IS_ERR(rt)) {
751 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
752 return ERR_PTR(-ENETUNREACH);
753 }
754 if (rt->dst.dev == dev) { /* is this necessary? */
755 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
756 ip_rt_put(rt);
757 return ERR_PTR(-ELOOP);
758 }
759 if (use_cache)
760 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
761 return rt;
762 }
763
764 #if IS_ENABLED(CONFIG_IPV6)
765 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
766 struct net_device *dev,
767 struct geneve_sock *gs6,
768 struct flowi6 *fl6,
769 const struct ip_tunnel_info *info)
770 {
771 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
772 struct geneve_dev *geneve = netdev_priv(dev);
773 struct dst_entry *dst = NULL;
774 struct dst_cache *dst_cache;
775 __u8 prio;
776
777 if (!gs6)
778 return ERR_PTR(-EIO);
779
780 memset(fl6, 0, sizeof(*fl6));
781 fl6->flowi6_mark = skb->mark;
782 fl6->flowi6_proto = IPPROTO_UDP;
783 fl6->daddr = info->key.u.ipv6.dst;
784 fl6->saddr = info->key.u.ipv6.src;
785 prio = info->key.tos;
786 if ((prio == 1) && !geneve->collect_md) {
787 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
788 use_cache = false;
789 }
790
791 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
792 info->key.label);
793 dst_cache = (struct dst_cache *)&info->dst_cache;
794 if (use_cache) {
795 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
796 if (dst)
797 return dst;
798 }
799 if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
800 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
801 return ERR_PTR(-ENETUNREACH);
802 }
803 if (dst->dev == dev) { /* is this necessary? */
804 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
805 dst_release(dst);
806 return ERR_PTR(-ELOOP);
807 }
808
809 if (use_cache)
810 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
811 return dst;
812 }
813 #endif
814
815 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
816 struct geneve_dev *geneve,
817 const struct ip_tunnel_info *info)
818 {
819 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
820 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
821 const struct ip_tunnel_key *key = &info->key;
822 struct rtable *rt;
823 struct flowi4 fl4;
824 __u8 tos, ttl;
825 __be16 sport;
826 __be16 df;
827 int err;
828
829 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info);
830 if (IS_ERR(rt))
831 return PTR_ERR(rt);
832
833 if (skb_dst(skb)) {
834 int mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr) -
835 GENEVE_BASE_HLEN - info->options_len - 14;
836
837 skb_dst_update_pmtu(skb, mtu);
838 }
839
840 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
841 if (geneve->collect_md) {
842 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
843 ttl = key->ttl;
844 } else {
845 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
846 ttl = key->ttl ? : ip4_dst_hoplimit(&rt->dst);
847 }
848 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
849
850 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
851 if (unlikely(err))
852 return err;
853
854 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
855 tos, ttl, df, sport, geneve->info.key.tp_dst,
856 !net_eq(geneve->net, dev_net(geneve->dev)),
857 !(info->key.tun_flags & TUNNEL_CSUM));
858 return 0;
859 }
860
861 #if IS_ENABLED(CONFIG_IPV6)
862 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
863 struct geneve_dev *geneve,
864 const struct ip_tunnel_info *info)
865 {
866 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
867 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
868 const struct ip_tunnel_key *key = &info->key;
869 struct dst_entry *dst = NULL;
870 struct flowi6 fl6;
871 __u8 prio, ttl;
872 __be16 sport;
873 int err;
874
875 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info);
876 if (IS_ERR(dst))
877 return PTR_ERR(dst);
878
879 if (skb_dst(skb)) {
880 int mtu = dst_mtu(dst) - sizeof(struct ipv6hdr) -
881 GENEVE_BASE_HLEN - info->options_len - 14;
882
883 skb_dst_update_pmtu(skb, mtu);
884 }
885
886 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
887 if (geneve->collect_md) {
888 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
889 ttl = key->ttl;
890 } else {
891 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
892 ip_hdr(skb), skb);
893 ttl = key->ttl ? : ip6_dst_hoplimit(dst);
894 }
895 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
896 if (unlikely(err))
897 return err;
898
899 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
900 &fl6.saddr, &fl6.daddr, prio, ttl,
901 info->key.label, sport, geneve->info.key.tp_dst,
902 !(info->key.tun_flags & TUNNEL_CSUM));
903 return 0;
904 }
905 #endif
906
907 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
908 {
909 struct geneve_dev *geneve = netdev_priv(dev);
910 struct ip_tunnel_info *info = NULL;
911 int err;
912
913 if (geneve->collect_md) {
914 info = skb_tunnel_info(skb);
915 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
916 err = -EINVAL;
917 netdev_dbg(dev, "no tunnel metadata\n");
918 goto tx_error;
919 }
920 } else {
921 info = &geneve->info;
922 }
923
924 rcu_read_lock();
925 #if IS_ENABLED(CONFIG_IPV6)
926 if (info->mode & IP_TUNNEL_INFO_IPV6)
927 err = geneve6_xmit_skb(skb, dev, geneve, info);
928 else
929 #endif
930 err = geneve_xmit_skb(skb, dev, geneve, info);
931 rcu_read_unlock();
932
933 if (likely(!err))
934 return NETDEV_TX_OK;
935 tx_error:
936 dev_kfree_skb(skb);
937
938 if (err == -ELOOP)
939 dev->stats.collisions++;
940 else if (err == -ENETUNREACH)
941 dev->stats.tx_carrier_errors++;
942
943 dev->stats.tx_errors++;
944 return NETDEV_TX_OK;
945 }
946
947 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
948 {
949 /* Only possible if called internally, ndo_change_mtu path's new_mtu
950 * is guaranteed to be between dev->min_mtu and dev->max_mtu.
951 */
952 if (new_mtu > dev->max_mtu)
953 new_mtu = dev->max_mtu;
954
955 dev->mtu = new_mtu;
956 return 0;
957 }
958
959 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
960 {
961 struct ip_tunnel_info *info = skb_tunnel_info(skb);
962 struct geneve_dev *geneve = netdev_priv(dev);
963
964 if (ip_tunnel_info_af(info) == AF_INET) {
965 struct rtable *rt;
966 struct flowi4 fl4;
967 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
968
969 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info);
970 if (IS_ERR(rt))
971 return PTR_ERR(rt);
972
973 ip_rt_put(rt);
974 info->key.u.ipv4.src = fl4.saddr;
975 #if IS_ENABLED(CONFIG_IPV6)
976 } else if (ip_tunnel_info_af(info) == AF_INET6) {
977 struct dst_entry *dst;
978 struct flowi6 fl6;
979 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
980
981 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info);
982 if (IS_ERR(dst))
983 return PTR_ERR(dst);
984
985 dst_release(dst);
986 info->key.u.ipv6.src = fl6.saddr;
987 #endif
988 } else {
989 return -EINVAL;
990 }
991
992 info->key.tp_src = udp_flow_src_port(geneve->net, skb,
993 1, USHRT_MAX, true);
994 info->key.tp_dst = geneve->info.key.tp_dst;
995 return 0;
996 }
997
998 static const struct net_device_ops geneve_netdev_ops = {
999 .ndo_init = geneve_init,
1000 .ndo_uninit = geneve_uninit,
1001 .ndo_open = geneve_open,
1002 .ndo_stop = geneve_stop,
1003 .ndo_start_xmit = geneve_xmit,
1004 .ndo_get_stats64 = ip_tunnel_get_stats64,
1005 .ndo_change_mtu = geneve_change_mtu,
1006 .ndo_validate_addr = eth_validate_addr,
1007 .ndo_set_mac_address = eth_mac_addr,
1008 .ndo_fill_metadata_dst = geneve_fill_metadata_dst,
1009 };
1010
1011 static void geneve_get_drvinfo(struct net_device *dev,
1012 struct ethtool_drvinfo *drvinfo)
1013 {
1014 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1015 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1016 }
1017
1018 static const struct ethtool_ops geneve_ethtool_ops = {
1019 .get_drvinfo = geneve_get_drvinfo,
1020 .get_link = ethtool_op_get_link,
1021 };
1022
1023 /* Info for udev, that this is a virtual tunnel endpoint */
1024 static struct device_type geneve_type = {
1025 .name = "geneve",
1026 };
1027
1028 /* Calls the ndo_udp_tunnel_add of the caller in order to
1029 * supply the listening GENEVE udp ports. Callers are expected
1030 * to implement the ndo_udp_tunnel_add.
1031 */
1032 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1033 {
1034 struct net *net = dev_net(dev);
1035 struct geneve_net *gn = net_generic(net, geneve_net_id);
1036 struct geneve_sock *gs;
1037
1038 rcu_read_lock();
1039 list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1040 if (push) {
1041 udp_tunnel_push_rx_port(dev, gs->sock,
1042 UDP_TUNNEL_TYPE_GENEVE);
1043 } else {
1044 udp_tunnel_drop_rx_port(dev, gs->sock,
1045 UDP_TUNNEL_TYPE_GENEVE);
1046 }
1047 }
1048 rcu_read_unlock();
1049 }
1050
1051 /* Initialize the device structure. */
1052 static void geneve_setup(struct net_device *dev)
1053 {
1054 ether_setup(dev);
1055
1056 dev->netdev_ops = &geneve_netdev_ops;
1057 dev->ethtool_ops = &geneve_ethtool_ops;
1058 dev->needs_free_netdev = true;
1059
1060 SET_NETDEV_DEVTYPE(dev, &geneve_type);
1061
1062 dev->features |= NETIF_F_LLTX;
1063 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1064 dev->features |= NETIF_F_RXCSUM;
1065 dev->features |= NETIF_F_GSO_SOFTWARE;
1066
1067 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1068 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1069
1070 /* MTU range: 68 - (something less than 65535) */
1071 dev->min_mtu = ETH_MIN_MTU;
1072 /* The max_mtu calculation does not take account of GENEVE
1073 * options, to avoid excluding potentially valid
1074 * configurations. This will be further reduced by IPvX hdr size.
1075 */
1076 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1077
1078 netif_keep_dst(dev);
1079 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1080 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1081 eth_hw_addr_random(dev);
1082 }
1083
1084 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1085 [IFLA_GENEVE_ID] = { .type = NLA_U32 },
1086 [IFLA_GENEVE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1087 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) },
1088 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
1089 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
1090 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 },
1091 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
1092 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
1093 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 },
1094 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1095 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1096 };
1097
1098 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1099 struct netlink_ext_ack *extack)
1100 {
1101 if (tb[IFLA_ADDRESS]) {
1102 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1103 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1104 "Provided link layer address is not Ethernet");
1105 return -EINVAL;
1106 }
1107
1108 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1109 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1110 "Provided Ethernet address is not unicast");
1111 return -EADDRNOTAVAIL;
1112 }
1113 }
1114
1115 if (!data) {
1116 NL_SET_ERR_MSG(extack,
1117 "Not enough attributes provided to perform the operation");
1118 return -EINVAL;
1119 }
1120
1121 if (data[IFLA_GENEVE_ID]) {
1122 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1123
1124 if (vni >= GENEVE_N_VID) {
1125 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1126 "Geneve ID must be lower than 16777216");
1127 return -ERANGE;
1128 }
1129 }
1130
1131 return 0;
1132 }
1133
1134 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1135 const struct ip_tunnel_info *info,
1136 bool *tun_on_same_port,
1137 bool *tun_collect_md)
1138 {
1139 struct geneve_dev *geneve, *t = NULL;
1140
1141 *tun_on_same_port = false;
1142 *tun_collect_md = false;
1143 list_for_each_entry(geneve, &gn->geneve_list, next) {
1144 if (info->key.tp_dst == geneve->info.key.tp_dst) {
1145 *tun_collect_md = geneve->collect_md;
1146 *tun_on_same_port = true;
1147 }
1148 if (info->key.tun_id == geneve->info.key.tun_id &&
1149 info->key.tp_dst == geneve->info.key.tp_dst &&
1150 !memcmp(&info->key.u, &geneve->info.key.u, sizeof(info->key.u)))
1151 t = geneve;
1152 }
1153 return t;
1154 }
1155
1156 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1157 {
1158 return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1159 info->key.ttl || info->key.label || info->key.tp_src ||
1160 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1161 }
1162
1163 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1164 struct ip_tunnel_info *b)
1165 {
1166 if (ip_tunnel_info_af(a) == AF_INET)
1167 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1168 else
1169 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1170 }
1171
1172 static int geneve_configure(struct net *net, struct net_device *dev,
1173 struct netlink_ext_ack *extack,
1174 const struct ip_tunnel_info *info,
1175 bool metadata, bool ipv6_rx_csum)
1176 {
1177 struct geneve_net *gn = net_generic(net, geneve_net_id);
1178 struct geneve_dev *t, *geneve = netdev_priv(dev);
1179 bool tun_collect_md, tun_on_same_port;
1180 int err, encap_len;
1181
1182 if (metadata && !is_tnl_info_zero(info)) {
1183 NL_SET_ERR_MSG(extack,
1184 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1185 return -EINVAL;
1186 }
1187
1188 geneve->net = net;
1189 geneve->dev = dev;
1190
1191 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1192 if (t)
1193 return -EBUSY;
1194
1195 /* make enough headroom for basic scenario */
1196 encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1197 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1198 encap_len += sizeof(struct iphdr);
1199 dev->max_mtu -= sizeof(struct iphdr);
1200 } else {
1201 encap_len += sizeof(struct ipv6hdr);
1202 dev->max_mtu -= sizeof(struct ipv6hdr);
1203 }
1204 dev->needed_headroom = encap_len + ETH_HLEN;
1205
1206 if (metadata) {
1207 if (tun_on_same_port) {
1208 NL_SET_ERR_MSG(extack,
1209 "There can be only one externally controlled device on a destination port");
1210 return -EPERM;
1211 }
1212 } else {
1213 if (tun_collect_md) {
1214 NL_SET_ERR_MSG(extack,
1215 "There already exists an externally controlled device on this destination port");
1216 return -EPERM;
1217 }
1218 }
1219
1220 dst_cache_reset(&geneve->info.dst_cache);
1221 geneve->info = *info;
1222 geneve->collect_md = metadata;
1223 geneve->use_udp6_rx_checksums = ipv6_rx_csum;
1224
1225 err = register_netdevice(dev);
1226 if (err)
1227 return err;
1228
1229 list_add(&geneve->next, &gn->geneve_list);
1230 return 0;
1231 }
1232
1233 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1234 {
1235 memset(info, 0, sizeof(*info));
1236 info->key.tp_dst = htons(dst_port);
1237 }
1238
1239 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1240 struct netlink_ext_ack *extack,
1241 struct ip_tunnel_info *info, bool *metadata,
1242 bool *use_udp6_rx_checksums, bool changelink)
1243 {
1244 int attrtype;
1245
1246 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1247 NL_SET_ERR_MSG(extack,
1248 "Cannot specify both IPv4 and IPv6 Remote addresses");
1249 return -EINVAL;
1250 }
1251
1252 if (data[IFLA_GENEVE_REMOTE]) {
1253 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1254 attrtype = IFLA_GENEVE_REMOTE;
1255 goto change_notsup;
1256 }
1257
1258 info->key.u.ipv4.dst =
1259 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1260
1261 if (IN_MULTICAST(ntohl(info->key.u.ipv4.dst))) {
1262 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1263 "Remote IPv4 address cannot be Multicast");
1264 return -EINVAL;
1265 }
1266 }
1267
1268 if (data[IFLA_GENEVE_REMOTE6]) {
1269 #if IS_ENABLED(CONFIG_IPV6)
1270 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1271 attrtype = IFLA_GENEVE_REMOTE6;
1272 goto change_notsup;
1273 }
1274
1275 info->mode = IP_TUNNEL_INFO_IPV6;
1276 info->key.u.ipv6.dst =
1277 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1278
1279 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1280 IPV6_ADDR_LINKLOCAL) {
1281 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1282 "Remote IPv6 address cannot be link-local");
1283 return -EINVAL;
1284 }
1285 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1286 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1287 "Remote IPv6 address cannot be Multicast");
1288 return -EINVAL;
1289 }
1290 info->key.tun_flags |= TUNNEL_CSUM;
1291 *use_udp6_rx_checksums = true;
1292 #else
1293 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1294 "IPv6 support not enabled in the kernel");
1295 return -EPFNOSUPPORT;
1296 #endif
1297 }
1298
1299 if (data[IFLA_GENEVE_ID]) {
1300 __u32 vni;
1301 __u8 tvni[3];
1302 __be64 tunid;
1303
1304 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1305 tvni[0] = (vni & 0x00ff0000) >> 16;
1306 tvni[1] = (vni & 0x0000ff00) >> 8;
1307 tvni[2] = vni & 0x000000ff;
1308
1309 tunid = vni_to_tunnel_id(tvni);
1310 if (changelink && (tunid != info->key.tun_id)) {
1311 attrtype = IFLA_GENEVE_ID;
1312 goto change_notsup;
1313 }
1314 info->key.tun_id = tunid;
1315 }
1316
1317 if (data[IFLA_GENEVE_TTL])
1318 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1319
1320 if (data[IFLA_GENEVE_TOS])
1321 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1322
1323 if (data[IFLA_GENEVE_LABEL]) {
1324 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1325 IPV6_FLOWLABEL_MASK;
1326 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1327 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1328 "Label attribute only applies for IPv6 Geneve devices");
1329 return -EINVAL;
1330 }
1331 }
1332
1333 if (data[IFLA_GENEVE_PORT]) {
1334 if (changelink) {
1335 attrtype = IFLA_GENEVE_PORT;
1336 goto change_notsup;
1337 }
1338 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1339 }
1340
1341 if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1342 if (changelink) {
1343 attrtype = IFLA_GENEVE_COLLECT_METADATA;
1344 goto change_notsup;
1345 }
1346 *metadata = true;
1347 }
1348
1349 if (data[IFLA_GENEVE_UDP_CSUM]) {
1350 if (changelink) {
1351 attrtype = IFLA_GENEVE_UDP_CSUM;
1352 goto change_notsup;
1353 }
1354 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1355 info->key.tun_flags |= TUNNEL_CSUM;
1356 }
1357
1358 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1359 #if IS_ENABLED(CONFIG_IPV6)
1360 if (changelink) {
1361 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1362 goto change_notsup;
1363 }
1364 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1365 info->key.tun_flags &= ~TUNNEL_CSUM;
1366 #else
1367 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1368 "IPv6 support not enabled in the kernel");
1369 return -EPFNOSUPPORT;
1370 #endif
1371 }
1372
1373 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1374 #if IS_ENABLED(CONFIG_IPV6)
1375 if (changelink) {
1376 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1377 goto change_notsup;
1378 }
1379 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1380 *use_udp6_rx_checksums = false;
1381 #else
1382 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1383 "IPv6 support not enabled in the kernel");
1384 return -EPFNOSUPPORT;
1385 #endif
1386 }
1387
1388 return 0;
1389 change_notsup:
1390 NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1391 "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1392 return -EOPNOTSUPP;
1393 }
1394
1395 static int geneve_newlink(struct net *net, struct net_device *dev,
1396 struct nlattr *tb[], struct nlattr *data[],
1397 struct netlink_ext_ack *extack)
1398 {
1399 bool use_udp6_rx_checksums = false;
1400 struct ip_tunnel_info info;
1401 bool metadata = false;
1402 int err;
1403
1404 init_tnl_info(&info, GENEVE_UDP_PORT);
1405 err = geneve_nl2info(tb, data, extack, &info, &metadata,
1406 &use_udp6_rx_checksums, false);
1407 if (err)
1408 return err;
1409
1410 return geneve_configure(net, dev, extack, &info, metadata,
1411 use_udp6_rx_checksums);
1412 }
1413
1414 /* Quiesces the geneve device data path for both TX and RX.
1415 *
1416 * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1417 * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1418 * to complete for the existing set of in-flight packets to be transmitted,
1419 * then we would have quiesced the transmit data path. All the future packets
1420 * will get dropped until we unquiesce the data path.
1421 *
1422 * On receive geneve dereference the geneve_sock stashed in the socket. So,
1423 * if we set that to NULL under RCU and wait for synchronize_net() to
1424 * complete, then we would have quiesced the receive data path.
1425 */
1426 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1427 struct geneve_sock **gs6)
1428 {
1429 *gs4 = rtnl_dereference(geneve->sock4);
1430 rcu_assign_pointer(geneve->sock4, NULL);
1431 if (*gs4)
1432 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1433 #if IS_ENABLED(CONFIG_IPV6)
1434 *gs6 = rtnl_dereference(geneve->sock6);
1435 rcu_assign_pointer(geneve->sock6, NULL);
1436 if (*gs6)
1437 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1438 #else
1439 *gs6 = NULL;
1440 #endif
1441 synchronize_net();
1442 }
1443
1444 /* Resumes the geneve device data path for both TX and RX. */
1445 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1446 struct geneve_sock __maybe_unused *gs6)
1447 {
1448 rcu_assign_pointer(geneve->sock4, gs4);
1449 if (gs4)
1450 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1451 #if IS_ENABLED(CONFIG_IPV6)
1452 rcu_assign_pointer(geneve->sock6, gs6);
1453 if (gs6)
1454 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1455 #endif
1456 synchronize_net();
1457 }
1458
1459 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1460 struct nlattr *data[],
1461 struct netlink_ext_ack *extack)
1462 {
1463 struct geneve_dev *geneve = netdev_priv(dev);
1464 struct geneve_sock *gs4, *gs6;
1465 struct ip_tunnel_info info;
1466 bool metadata;
1467 bool use_udp6_rx_checksums;
1468 int err;
1469
1470 /* If the geneve device is configured for metadata (or externally
1471 * controlled, for example, OVS), then nothing can be changed.
1472 */
1473 if (geneve->collect_md)
1474 return -EOPNOTSUPP;
1475
1476 /* Start with the existing info. */
1477 memcpy(&info, &geneve->info, sizeof(info));
1478 metadata = geneve->collect_md;
1479 use_udp6_rx_checksums = geneve->use_udp6_rx_checksums;
1480 err = geneve_nl2info(tb, data, extack, &info, &metadata,
1481 &use_udp6_rx_checksums, true);
1482 if (err)
1483 return err;
1484
1485 if (!geneve_dst_addr_equal(&geneve->info, &info))
1486 dst_cache_reset(&info.dst_cache);
1487
1488 geneve_quiesce(geneve, &gs4, &gs6);
1489 geneve->info = info;
1490 geneve->collect_md = metadata;
1491 geneve->use_udp6_rx_checksums = use_udp6_rx_checksums;
1492 geneve_unquiesce(geneve, gs4, gs6);
1493
1494 return 0;
1495 }
1496
1497 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1498 {
1499 struct geneve_dev *geneve = netdev_priv(dev);
1500
1501 list_del(&geneve->next);
1502 unregister_netdevice_queue(dev, head);
1503 }
1504
1505 static size_t geneve_get_size(const struct net_device *dev)
1506 {
1507 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
1508 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1509 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
1510 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
1511 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */
1512 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
1513 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1514 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1515 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1516 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1517 0;
1518 }
1519
1520 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1521 {
1522 struct geneve_dev *geneve = netdev_priv(dev);
1523 struct ip_tunnel_info *info = &geneve->info;
1524 bool metadata = geneve->collect_md;
1525 __u8 tmp_vni[3];
1526 __u32 vni;
1527
1528 tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1529 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1530 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1531 goto nla_put_failure;
1532
1533 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1534 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1535 info->key.u.ipv4.dst))
1536 goto nla_put_failure;
1537 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1538 !!(info->key.tun_flags & TUNNEL_CSUM)))
1539 goto nla_put_failure;
1540
1541 #if IS_ENABLED(CONFIG_IPV6)
1542 } else if (!metadata) {
1543 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1544 &info->key.u.ipv6.dst))
1545 goto nla_put_failure;
1546 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1547 !(info->key.tun_flags & TUNNEL_CSUM)))
1548 goto nla_put_failure;
1549 #endif
1550 }
1551
1552 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1553 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1554 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1555 goto nla_put_failure;
1556
1557 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1558 goto nla_put_failure;
1559
1560 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1561 goto nla_put_failure;
1562
1563 #if IS_ENABLED(CONFIG_IPV6)
1564 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1565 !geneve->use_udp6_rx_checksums))
1566 goto nla_put_failure;
1567 #endif
1568
1569 return 0;
1570
1571 nla_put_failure:
1572 return -EMSGSIZE;
1573 }
1574
1575 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1576 .kind = "geneve",
1577 .maxtype = IFLA_GENEVE_MAX,
1578 .policy = geneve_policy,
1579 .priv_size = sizeof(struct geneve_dev),
1580 .setup = geneve_setup,
1581 .validate = geneve_validate,
1582 .newlink = geneve_newlink,
1583 .changelink = geneve_changelink,
1584 .dellink = geneve_dellink,
1585 .get_size = geneve_get_size,
1586 .fill_info = geneve_fill_info,
1587 };
1588
1589 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1590 u8 name_assign_type, u16 dst_port)
1591 {
1592 struct nlattr *tb[IFLA_MAX + 1];
1593 struct ip_tunnel_info info;
1594 struct net_device *dev;
1595 LIST_HEAD(list_kill);
1596 int err;
1597
1598 memset(tb, 0, sizeof(tb));
1599 dev = rtnl_create_link(net, name, name_assign_type,
1600 &geneve_link_ops, tb);
1601 if (IS_ERR(dev))
1602 return dev;
1603
1604 init_tnl_info(&info, dst_port);
1605 err = geneve_configure(net, dev, NULL, &info, true, true);
1606 if (err) {
1607 free_netdev(dev);
1608 return ERR_PTR(err);
1609 }
1610
1611 /* openvswitch users expect packet sizes to be unrestricted,
1612 * so set the largest MTU we can.
1613 */
1614 err = geneve_change_mtu(dev, IP_MAX_MTU);
1615 if (err)
1616 goto err;
1617
1618 err = rtnl_configure_link(dev, NULL);
1619 if (err < 0)
1620 goto err;
1621
1622 return dev;
1623 err:
1624 geneve_dellink(dev, &list_kill);
1625 unregister_netdevice_many(&list_kill);
1626 return ERR_PTR(err);
1627 }
1628 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1629
1630 static int geneve_netdevice_event(struct notifier_block *unused,
1631 unsigned long event, void *ptr)
1632 {
1633 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1634
1635 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1636 event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1637 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1638 } else if (event == NETDEV_UNREGISTER) {
1639 geneve_offload_rx_ports(dev, false);
1640 } else if (event == NETDEV_REGISTER) {
1641 geneve_offload_rx_ports(dev, true);
1642 }
1643
1644 return NOTIFY_DONE;
1645 }
1646
1647 static struct notifier_block geneve_notifier_block __read_mostly = {
1648 .notifier_call = geneve_netdevice_event,
1649 };
1650
1651 static __net_init int geneve_init_net(struct net *net)
1652 {
1653 struct geneve_net *gn = net_generic(net, geneve_net_id);
1654
1655 INIT_LIST_HEAD(&gn->geneve_list);
1656 INIT_LIST_HEAD(&gn->sock_list);
1657 return 0;
1658 }
1659
1660 static void __net_exit geneve_exit_net(struct net *net)
1661 {
1662 struct geneve_net *gn = net_generic(net, geneve_net_id);
1663 struct geneve_dev *geneve, *next;
1664 struct net_device *dev, *aux;
1665 LIST_HEAD(list);
1666
1667 rtnl_lock();
1668
1669 /* gather any geneve devices that were moved into this ns */
1670 for_each_netdev_safe(net, dev, aux)
1671 if (dev->rtnl_link_ops == &geneve_link_ops)
1672 unregister_netdevice_queue(dev, &list);
1673
1674 /* now gather any other geneve devices that were created in this ns */
1675 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1676 /* If geneve->dev is in the same netns, it was already added
1677 * to the list by the previous loop.
1678 */
1679 if (!net_eq(dev_net(geneve->dev), net))
1680 unregister_netdevice_queue(geneve->dev, &list);
1681 }
1682
1683 /* unregister the devices gathered above */
1684 unregister_netdevice_many(&list);
1685 rtnl_unlock();
1686 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1687 }
1688
1689 static struct pernet_operations geneve_net_ops = {
1690 .init = geneve_init_net,
1691 .exit = geneve_exit_net,
1692 .id = &geneve_net_id,
1693 .size = sizeof(struct geneve_net),
1694 };
1695
1696 static int __init geneve_init_module(void)
1697 {
1698 int rc;
1699
1700 rc = register_pernet_subsys(&geneve_net_ops);
1701 if (rc)
1702 goto out1;
1703
1704 rc = register_netdevice_notifier(&geneve_notifier_block);
1705 if (rc)
1706 goto out2;
1707
1708 rc = rtnl_link_register(&geneve_link_ops);
1709 if (rc)
1710 goto out3;
1711
1712 return 0;
1713 out3:
1714 unregister_netdevice_notifier(&geneve_notifier_block);
1715 out2:
1716 unregister_pernet_subsys(&geneve_net_ops);
1717 out1:
1718 return rc;
1719 }
1720 late_initcall(geneve_init_module);
1721
1722 static void __exit geneve_cleanup_module(void)
1723 {
1724 rtnl_link_unregister(&geneve_link_ops);
1725 unregister_netdevice_notifier(&geneve_notifier_block);
1726 unregister_pernet_subsys(&geneve_net_ops);
1727 }
1728 module_exit(geneve_cleanup_module);
1729
1730 MODULE_LICENSE("GPL");
1731 MODULE_VERSION(GENEVE_NETDEV_VER);
1732 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1733 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1734 MODULE_ALIAS_RTNL_LINK("geneve");
ubuntu-bionic-kernel mirror